This package is still under development. If you have any trouble running this code, please contact thomas.moreau.2010@gmail.com

DICOD

Package to run the experiments for the ICML paper DICOD: Distributed Convolutional Coordinate Descent for Convolutional Sparse Coding, ICML 2018, T. Moreau, L. Oudre, N. Vayatis.

Requirements

All the tests were done with python3.4. This package depends on the python library numpy, matplotlib, scipy, mpi4py, joblib and the libraries openMPI and fftw3. They can be installed with

sudo apt install libopenmpi-dev fftw-dev
pip install numpy matplotlib scipy mpi4py joblib

To install the package, first build it with the utility script ./build and then run pip install -e .

Usage

Figure 2 can be generated using

$ python main_dicod.py --met -K 25 -T 600 --timeout 7200 -d 10 --njobs 60 --hostfile hostfile --exp results

where hostfile is the configuration for the spawning of MPI processes.

host1 slots=32
host2 slots=8
...

Then the figures can be plotted using

$ python plot_dicod.py --met --dir save_exp/results